Systems and Methods for Brain Hemorrhage Classification in Medical Images Using an Artificial Intelligence Network

1. A method for identifying a condition in a medical image of a subject comprising: a) accessing a medical image of a subject with a computer system; b) generating a plurality of channel images from the medical image, each channel image being generated by applying at least one of a window setting and a level setting to the medical image based on a color, or color intensity; c) identifying a region-of-interest (ROI) in the medical image by processing the plurality of channel images with an artificial intelligence (AI) network implemented with a hardware processor and a memory of the computer system; d) labelling the ROI with an identification of a first brain hemorrhage condition in the ROI using an output of the AI network; and e) generating a report for a user identifying the first brain hemorrhage condition and including a confidence value associated with an identification of the first brain hemorrhage condition in the ROI.

2. The method of claim 1 wherein the first brain hemorrhage condition includes at least one of an intracranial hemorrhage (ICH), an intraventricular hemorrhage (IVH), a subarachnoid hemorrhage (SAH), an intra-parenchymal hemorrhage (IPH), an epidural hematoma (EDH), a subdural hematoma (SDH), or a bleed.

3. The method of claim 1 further comprising identifying a second ROI in the medical imaging by processing the medical image with the AI network, and labelling the second ROI with an identification of a second brain hemorrhage condition using an output of the AI network.

4. The method of claim 3 wherein generating the report includes displaying a first image slice in which the first brain hemorrhage condition is identified and displaying a second image slice in which the second brain hemorrhage condition is identified.

5. The method of claim 3 wherein generating the report includes displaying an image slice in which both the first brain hemorrhage condition and the second brain hemorrhage condition are identified, and wherein identifying the second condition includes a second confidence value associated with an identification of the second brain hemorrhage condition in the second ROI.

6. The method of claim 1 wherein the plurality of channel images are combined to form a combined image and inputting the plurality of channel images to the AI network comprises inputting the combined image to the AI network.

7. The method of claim 1 wherein each of the plurality of channel images is assigned a different color.

8. The method of claim 7 wherein each of the plurality of channel images is assigned the different color by mapping pixel intensities in each channel image to a different color scale.

9. The method of claim 8 wherein the plurality of channel images comprises a first channel image, a second channel image, and a third channel image, and wherein pixel intensities in the first channel image are mapped to a red color scale, pixel intensities in the second channel image are mapped to a green color scale, and pixel intensities in the third channel image are mapped to a blue color scale.

10. The method of claim 1 further comprising preprocessing the medical image prior to inputting the medical image to the AI network, wherein preprocessing the medical image includes reducing noise in the medical image.

11. The method of claim 1 wherein the medical image is an image volume comprising a plurality of image slices, and wherein the plurality of image slices are interpolated before inputting the medical image to the AI network.

12. The method of claim 11 wherein the plurality of image slices are interpolated to generate at least one new image slice between two of the plurality of image slices.

13. The method of claim 11 wherein the image volume is a three-dimensional image volume comprising a stack of spatially adjacent image slices.

14. The method of claim 11 wherein the image volume comprises a series of temporally adjacent image slices.

15. The method of claim 1 wherein the AI network is a convolutional neural network.

16. The method of claim 1 wherein the output of the AI network comprises a class activation map.

17. A system for identifying a brain hemorrhage condition in an image of a subject comprising: at least one hardware processor; a memory having stored thereon instructions that when executed by the at least one hardware processor cause the at least one hardware processor to perform steps comprising: a) accessing a medical image of a subject from the memory; b) generating a plurality of channel images from the medical image, each channel image being generated by applying at least one of a window setting and a level setting to the medical image based on a color, or color intensity; c) accessing a trained convolutional neural network (CNN) from the memory, the trained CNN having been trained on medical images labeled with one or more different brain hemorrhage conditions; d) generating a class activation map that indicates at least one brain hemorrhage condition in the medical image by processing the plurality of channel images with the AI network; e) generating a labeled image by labelling regions in the medical image associated with the at least one brain hemorrhage condition using the class activation map; f) producing at least one confidence value for each labeled region in the labeled image, each confidence value being associated with a confidence of each labeled region representing a corresponding at least one brain hemorrhage condition; and g) generating a display that depicts the labeled image and the at least one confidence value.

18. The system of claim 17 wherein the at least one brain hemorrhage condition includes at least one of an intracranial hemorrhage (ICH), an intraventricular hemorrhage (IVH), a subarachnoid hemorrhage (SAH), an intra-parenchymal hemorrhage (IPH), an epidural hematoma (EDH), a subdural hematoma (SDH), or a bleed.